New technology can reduce bias in loan underwriting

In the summer of 2021, I wrote that consumer finance in the U.S. could borrow a page from the field of medicine in order to help raise the “standard of care” in how we fairly underwrite, reduce bias and lend to the many millions of historically disadvantaged consumers in the country.

My argument rests on the notion that our expectations should rise as innovation drives new technological capabilities. The U.S. Treasury Department apparently agrees. Its recent report stating as much should profoundly impact consumer finance policy in the U.S.

On Nov. 16, 2022, the Treasury Department released a report entitled “Assessing Impacts of New Entrant Non-bank Firms on Competition in Consumer Finance Markets” and highlighted a primary recommendation in the accompanying press release: “To encourage consumer-beneficial innovation, regulators should support innovations in consumer credit underwriting designed to increase credit visibility, reduce bias, and prudently expand credit to underserved consumers.”

This is powerful recognition that it is time to raise the standard of care and that current innovations are the path forward in doing so.

The Treasury report notes that “[c]onventional credit underwriting, with its reliance on traditional automated scoring models (e.g., classic FICO) … may leave gaps … [and] can leave out or underscore creditworthy consumers simply because they lack established credit histories with mainstream lenders (e.g., mortgage lenders, credit cards), and may particularly impact borrowers that are young, low-income, or minority.” The report goes on to suggest that regulators — largely through guidance — should take steps to ensure that lenders have the confidence to incorporate modern underwriting technologies in a compliant manner.

But what specifically can regulators do to satisfy the Treasury’s call to action in raising the standard of care?

First, regulators should issue guidance that explicitly recognizes that certain artificial intelligence and machine learning models can be even more transparent and explainable than traditional underwriting approaches. This simple recognition would help to debunk a prevailing view that AI/ML necessarily means “black box” algorithms whose decisions cannot be explained. Such guidance would need to delve deeper into understanding the types of quantitative explainability methods available today and would require supervised lenders to utilize such methodologies when assessing their models. FinRegLab’s April 2022 report is a great starting point for that analysis. 

Second, regulators should ensure that supervised lenders create documentation showing that they rigorously tested their algorithms to make sure there was not a less discriminatory alternative. Today, guidance on how such searching should be conducted is limited, at best. Guidance that increases the standard of care while leaving room for innovation and further advancement will be critical. 

Third, regulators can issue guidance designed to encourage lenders to upgrade existing models with fairer ones by agreeing not to use the adoption of a new model as evidence of prior wrongdoing. In other areas of law this is called the doctrine of “subsequent remedial measures.” Without it, people may be reluctant to make improvements for fear that such improvements would be used against them. This sort of guidance would eliminate those perverse incentives. Such guidance should, of course, include safeguards to ensure that it is not used as a pass for prior discrimination. But the need for guardrails is not an argument against building a better road. 

There are other ways that regulators might help implement the Treasury’s recommendations. For instance, they might look for ways to hold legacy score providers to similar standards of fairness and transparency that they (rightly) demand of new entrants. While legacy providers’ stranglehold on the market may put pressure on regulators to effectively grandfather-in such legacy models, making them conduct robust fair-lending analysis and provide detailed documentation to industry shouldn’t be too much to ask. 

We did not get to safer cars, elevators, spacecraft or medical procedures without innovation and a clear recognition that progress is possible. And we will not get to a fairer financial future for underserved Americans without similar resolve — a point the U.S. Treasury appears to recognize. A standard of care framework is one way to promote and even require innovation, while minimizing risks to consumers. There are probably others as well. But now it’s time for the federal financial regulators to implement the policies the Treasury has called for.